Enabling remote work and new scenarios with Microsoft 365 E3

by Contributed | Feb 18, 2021 | Technology

This article is contributed. See the original author and article here.

IT decision makers are being asked to ensure their businesses are secure and their end users are productive – regardless of where they are working. To support this need, our customers are looking to accelerate their growth in the cloud with Microsoft 365. The Microsoft 365 E3 plan provides the flexibility to adapt and thrive through societal and economic disruptions because it combines best-in-class productivity apps with core security and compliance capabilities. One of the most powerful tools in the Microsoft 365 toolkit is Microsoft Endpoint Manager.

In the last eighteen months, customers and partners have taken advantage of several product innovations to not only simplify the consumption of the latest technologies in their organizations, but also adapt to changing demands of remote work without missing a beat.

Stay protected when working remotely

 Here are some highlights of recent announcements, with even more coming very soon:

• Microsoft Tunnel virtual private network (VPN) capabilities, extending Conditional Access to workers to use Android and iOS devices to remotely access on-premises data – a key principle of Zero Trust security.


• Broadened endpoint management, including adding support for virtual desktops and kiosks and shared devices, in addition to Windows PCs, Macs, and mobile devices.


• Expanded MacOS management to deliver more apps securely, collect better telemetry to improve management decisions, and improve security with enterprise Single Sign-On (SSO) and file encryption.


• Endpoint analytics to help you understand how performance and health issues with your organization’s hardware and software may be impacting your end-user’s productivity. Further, Endpoint analytics connects insights from Productivity Score with Microsoft Endpoint Manager to ensure everyone can do their best work with Microsoft 365.


• Unified endpoint security management that reduces the friction between end-user computing teams and security operations by providing deep integration between Microsoft Endpoint Manager and Microsoft Defender for Endpoint, as well as enabling specialized security roles for security managers to view and manage their day-to-day tasks within the Configuration Manager Console.


• Modernize your IT by bringing intelligent security value to all endpoints, whether managed on-premises or in the cloud.

Microsoft 365 helps foster a culture of collaboration with connected experiences while protecting company data, increasing employee satisfaction, and optimizing the administration of security and management.

Given these significant investments in Microsoft Endpoint Manger and Microsoft Enterprise Mobility + Security (EMS) over the past 18 months – with even more to come in the coming weeks – we plan to increase the price of these standalone products: EMS E3 from $9 to $11, and Intune standalone from $6 to $8 effective July 1, 2021. The price for Microsoft 365 E3 will not change.

Continue the conversation by joining us in the Microsoft 365 Tech Community! Whether you have product questions or just want to stay informed with the latest updates on new releases, tools, and blogs, Microsoft 365 Tech Community is your go-to resource to stay connected!

Enabling productivity for everyone with Microsoft 365 Apps for enterprise

by Contributed | Feb 18, 2021 | Technology

This article is contributed. See the original author and article here.

Enabling productivity for everyone with Microsoft 365 Apps for enterprise  

A huge benefit to being in the cloud with Microsoft 365 is having the flexibility to provide employees working across different roles and locations with the latest features and updates—a critical task for any IT organization. That’s why we’re excited to announce two new features for Microsoft 365 Apps for enterprise—extended offline access and device-based subscriptions—that remove critical blockers and enable customers to deploy Microsoft 365 across their entire environment, streamlining deployment and administration. These capabilities will help you ensure that employees who are offline for months at a time or rely on shared devices and workstations can benefit from the same user experience and stay productive and secure no matter where they’re working. 

Enabling extended offline access scenarios

To ensure Microsoft 365 Apps stays up to date, devices must connect to the internet at least once every 30 days. However, we’re aware that in industries, including government, oil and gas, manufacturing, agriculture, and scientific research, some people work in secure or remote environments where they have limited or no internet connectivity for longer periods of time.  

To address this, we’re now providing extended offline access* to enable devices to stay activated without connecting to the internet for up to 180 days. Workers in secure or remote environments who are offline for long periods of time can continue using Microsoft 365 Apps to stay productive on-the-job without worrying about being cut off from the tools they need most after 30 days.  

Setting up and running Microsoft 365 Apps for offline use 

For organizations with workers who need to run Microsoft 365 Apps offline for an extended period, IT administrators can enable extended offline access when they install Microsoft 365 Apps on a device. The worker signs into Windows with their Microsoft 365 account by viewing the expiration date that appears in a Product Information window on their device. After that, the worker can continue using Office with no internet for up to 180 days. Fifteen (15) days before offline access expires, they will receive an in-app notification. At that point, the worker can either reconnect the device to the internet before the expiration date or the IT administrator can generate a license in the Office portal from a second, connected device and copy the license to the other device.  

In order to enable Extended Offline Access on a device, IT admins need to deploy the group policy on that device

Setting up device-based subscriptions for shared devices

For organizations, whose employees are mostly information workers, user-based licensing for Microsoft 365 covers most use case scenarios. But for organizations in industries, such as manufacturing, agriculture, healthcare, retail and hospitality, many employees may share one device. In those cases, users that rely on shared devices have not been able to have access to the latest and most secure productivity tools that are available on desktop, to address this scenario we’re introducing device-based subscriptions for Microsoft 365 Apps.  

Assigning licenses to devices in Microsoft 365 admin center.

Having device-based subscriptions for Microsoft 365 Apps enables you to extend coverage to commonly used devices on loading docks, at nurses’ stations, on the manufacturing floor, or in a breakroom. Because the license is assigned to the device, workers aren’t required to have their own Azure Active Directory identity. Workers can sign into the device as many times as needed and access all Microsoft 365 Apps, including Excel, OneNote, Outlook, PowerPoint, Publisher, and Word.

To deploy a device-based subscription, you simply purchase the required number of Microsoft 365 licenses and assign a license to a device group in the Microsoft 365 admin center. To enable this functionality on a device, use the group policy for currently installed devices and/or the configuration.xml attribute. 

Learn more about device-based licensing for Microsoft 365 Apps for enterprise here.

*Eligible customers should contact their Microsoft account representative to determine if extended offline access for Microsoft 365 Apps for enterprise is the right solution for them.

Continue the conversation by joining us in the Microsoft 365 Tech Community! Whether you have product questions or just want to stay informed with the latest updates on new releases, tools, and blogs, Microsoft 365 Tech Community is your go-to resource to stay connected!

Responsible Machine Learning with Error Analysis

by Contributed | Feb 18, 2021 | Technology

This article is contributed. See the original author and article here.

Website: ErrorAnalysis.ai

Github repository: https://github.com/microsoft/responsible-ai-widgets/

 

Machine Learning (ML) teams who deploy models in the real world often face the challenges of conducting rigorous performance evaluation and testing for ML models. How often do we read claims such as “Model X is 90% on a given benchmark.” and wonder what does this claim mean for practical usage of the model? In practice, teams are well aware that model accuracy may not be uniform across subgroups of data and that there might exist input conditions for which the model fails more often. Often, such failures may cause direct consequences related to lack of reliability and safety, unfairness, or more broadly lack of trust in machine learning altogether. For instance, when a traffic sign detector does not operate well in certain daylight conditions or for unexpected inputs, even though the overall accuracy of the model may be high, it is still important for the development team to know ahead of time about the fact that the model may not be as reliable in such situations.

 

 

Figure 1 – Error Analysis moves away from aggregate accuracy metrics, exposes the distribution of errors to developers in a transparent way, and enables them to identify & diagnose errors efficiently.

 

While there exist several problems with current model assessment practices, one of the most obvious is the usage of aggregate metrics to score models on a whole benchmark. It is difficult to convey a detailed story on model behavior with a single number and yet most of the research and leaderboards operate on single scores. At the same time, there may exist several dimensions of the input feature space that a practitioner may be interested in taking a deep dive and ask questions such as “What happens to the accuracy of the recognition model in a self-driving car when it is dark and snowing outside?” or “Does the loan approval model perform similarly for population cohorts across ethnicity, gender, age, and education?”. Navigating the terrain of failures along multiple potential dimensions like the above can be challenging. In addition, in the longer term, when models are updated and re-deployed frequently upon new data evidence or scientific progress, teams also need to continuously track and monitor model behavior so that updates do not introduce new mistakes and therefore break user trust.

 

To address these problems, practitioners often have to create custom infrastructure, which is tedious and time-consuming. To accelerate rigorous ML development, in this blog you will learn how to use the Error Analysis tool for:

• Getting a deep understanding of how failure is distributed for a model.


• Debugging ML errors with active data exploration and interpretability techniques.

The Error Analysis toolkit is integrated within the Responsible AI Widgets OSS repository, our starting point to provide a set of integrated tools to the open source community and ML practitioners. Not only a contribution to the OSS RAI community, but practitioners can also leverage these assessment tools in Azure Machine Learning, including Fairlearn & InterpretML and now Error Analysis in mid 2021.

 

If you are interested in learning more about training model updates that remain backward compatible with their previous selves by minimizing regress and new errors, you can also check out our most recent open source library and tool BackwardCompatibilityML.

 

Prerequisites

To install the Responsible AI Widgets “raiwidgets” package, in your python environment simply run the following to install the raiwidgets package from pypi. If you do not have interpret-community already installed, you will also need to install this for supporting the generation of model explanations. 

pip install interpret-community
pip install raiwidgets

Alternatively, you can also clone the open source repository and build the code from scratch:

git clone https://github.com/microsoft/responsible-ai-widgets.git

You will need to install yarn and node to build the visualization code, and then you can run:

yarn install
yarn buildall

And install from the raiwidgets folder locally:

cd raiwidgets
pip install –e .

For more information see the contributing guide.

If you intend to run repository tests, in the raiwidgets folder of the repository run:

pip install -r requirements.txt

 

Getting started

This post illustrates the Error Analysis tool by using a binary classification task on income prediction (>50K, <50K). The model under inspection will be trained using the tabular UCI Census Income dataset, which contains both numerical and categorical features such as age, education, number of working hours, ethnicity, etc.

 

We can call the error analysis dashboard using the API below, which takes in an explanation object computed by one of the explainers from the interpret-community repository, the model or pipeline, a dataset and the corresponding labels (true_y parameter):

ErrorAnalysisDashboard(global_explanation, model, dataset=x_test, true_y=y_test)

For larger datasets, we can downsample the explanation to fewer rows but run error analysis on the full dataset.  We can provide the downsampled explanation, the model or pipeline, the full dataset, and then both the labels for the sampled explanation and the full dataset, as well as (optionally) the names of the categorical features:

ErrorAnalysisDashboard(global_explanation, model, dataset=X_test_original_full,true_y=y_test, categorical_features=categorical_features, true_y_dataset=y_test_full)

All screenshots below are generated using a LGBMClassifier with three estimators. You can directly run this example using the jupyter notebooks in our repository.

 

How Error Analysis works

 

1. Identification

Error Analysis starts with identifying the cohorts of data with a higher error rate versus the overall benchmark error rate. The dashboard allows for error exploration by using either an error heatmap or a decision tree guided by errors.

 

Error Heatmap for Error Identification

The view slices the data based on a one- or two-dimensional grid of input features. Users can choose the input features of interest for analysis. The heatmap visualizes cells with higher error with a darker red color to bring the user’s attention to regions with high error discrepancy. This is beneficial especially when the error themes are different in different partitions, which happens frequently in practice. In this error identification view, the analysis is highly guided by the users and their knowledge or hypotheses of what features might be most important for understanding failure.  

 

 

Figure 2 – While the overall error rate for the dataset is 23.65%, the heatmap reveals that the error rates are visibly higher, up to 83%, for individuals with higher education. Error rates are also higher for males vs. females.

 

Decision Tree for Error Identification

Very often, error patterns may be complex and involve more than one or two features. Therefore, it may be difficult for developers to explore all possible combinations of features to discover hidden data pockets with critical failure. To alleviate the burden, the binary tree visualization automatically partitions the benchmark data into interpretable subgroups, which have unexpectedly high or low error rates. In other words, the tree leverages the input features to maximally separate model error from success. For each node defining a data subgroup, users can investigate the following information:

• Error rate – a portion of instances in the node for which the model is incorrect. This is shown through the intensity of the red color.


• Error coverage – a portion of all errors that fall into the node. This is shown through the fill rate of the node.           


• Data representation – number of instances in the node. This is shown through the thickness of the incoming edge to the node along with the actual total number of instances in the node.

 

Figure 3 – Decision tree that aims at finding failure modes by separating error instances from success instances in the data. The hierarchical error pattern here shows that while the overall error rate is 23.65% for the dataset, it can be as high as 96.77% for individuals who are married, have a capital gain higher than 4401, and a number of education years higher than 12.

 

Cohort definition and manipulation

To specialize the analysis and allow for deep dives, both error identification views can be generated for any data cohort and not only for the whole benchmark. Cohorts are subgroups of data that the user may choose to save for later use if they wish to come back to those cohorts for future investigation. They can be defined and manipulated interactively either from the heatmap or the tree. They can also be carried over to the next diagnostical views on data exploration and model explanations.

 

 

Figure 4 – Creating a new cohort for further investigation that focuses on individuals who are married and have capital gain lower than 4401.

 

2. Diagnosis

After identifying cohorts with higher error rates, Error Analysis enables debugging and exploring these cohorts further. It is then possible to gain deeper insights about the model or the data through data exploration and model interpretability.

 

Debugging the data

Data Explorer: Users can explore dataset statistics and distributions by selecting different features and estimators along the two axes of the data explorer. They can further compare the subgroup data stats with other subgroups or the overall benchmark data. This view can for instance uncover if certain cohorts are underrepresented or if their feature distribution is significantly different from the overall data, hinting therefore to the potential existence of outliers or unusual covariate shift.

 

 

Figure 5 – In figure 1 and 2, we discovered that for individuals with a higher number of education years, the model has higher failure rates. When we look at how the data is distributed across the feature “education_num” we can see that a) there are fewer instances for individuals with more than 12 years of education, and b) for this cohort the distribution between lower income (blue) and higher income (orange) is very different than for other cohorts. In fact, for this cohort there exist more people who have an income higher than 50K, which is not true for the overall data.

 

Instance views: Beyond data statistics, sometimes it is useful to merely just observe the raw data along with labels in a tabular or tile form. Instance views provide this functionality and divide the instances into correct and incorrect tabs. By eyeballing the data, the developer can identify potential issues related to missing features or label noise.

 

Debugging the model

Model interpretability is a powerful means for extracting knowledge on how a model works. To extract this knowledge, Error Analysis relies on Microsoft’s InterpretML dashboard and library. The library is a prominent contribution in ML interpretability lead by Rich Caruana, Paul Koch, Harsha Nori, and Sam Jenkins.  

 

Global explanations

Feature Importance: Users can explore the top K important features that impact the overall model predictions (a.k.a. global explanation) for a selected subgroup of data or cohort. They can also compare feature importance values for different cohorts side by side. The information on feature importance or the ordering is useful for understanding whether the model is leveraging features that are necessary for the prediction or whether it is relying on spurious correlations. By contrasting explanations that are specific to the cohort with those for the whole benchmark, it is possible to understand whether the model behaves differently or in an unusual way for the selected cohort.

 

Dependence Plot: Users can see the relationship between the values of the selected feature to its corresponding feature importance values. This shows them how values of the selected feature impact model prediction.

 

 

Figure 6 – Global feature explanations for the income prediction model show that marital status and number of education years are the most important features globally. By clicking on each feature, it is possible to observe more granular dependencies. For example, marital statuses like “divorced”, “never married”, “separated”, or “widowed” contribute to model predictions for lower income (<50K). Marital status of “civil spouse” instead contributes to model predictions for higher income (>50K).

 

Local explanations

Global explanations approximate the overall model behavior. For focusing the debugging process on a given data instance, users can select any individual data points (with correct or incorrect predictions) from the tabular instance view to explore their local feature importance values (local explanation) and individual conditional expectation (ICE) plots.

 

Local Feature Importance: Users can investigate the top K (configurable K) important features for an individual prediction. Helps illustrate the local behavior of the underlying model on a specific data point.

 

Individual Conditional Expectation (ICE): Users can investigate how changing a feature value from a minimum value to a maximum value impacts the prediction on the selected data instance.

 

Perturbation Exploration (what-if analysis): Users can apply changes to feature values of the selected data point and observe resulting changes to the prediction. They can save their hypothetical what-if data points for further comparisons with other what-if or original data points.

 

 

Figure 7 – For this individual, the model outputs a wrong prediction, predicting that the individual earns less than 50K, while the opposite is true. With what-if explanations, it is possible to understand how the model would behave if one of the feature values changes. For instance, here we can see that if the individual were 10 years older (age changed from 32 to 42) the model would have made a correct prediction. While in the real world many of these features are not mutable, this sensitivity analysis is intended to further support practitioners with model understanding capabilities.

 

Other relevant tools

Error Analysis enables practitioners to identify and diagnose error patterns. The integration with model interpretability techniques testifies to the joint power of providing such tools together as part of the same platform. We are actively working towards integrating further considerations into the model assessment experience such as fairness and inclusion (via FairLearn) as well as backward compatibility during updates (via BackwardCompatibilityML).

 

Our team

The initial work on error analysis started with research investigations on methodologies for in-depth understanding and explanation of Machine Learning failures. Besmira Nushi, Ece Kamar, and Eric Horvitz at Microsoft Research are leading these efforts and continue to innovate with new techniques for debugging ML models. In the past year, our team was extended via a collaboration with the RAI tooling team in the Azure Machine Learning group as well as the Analysis Platform team in Microsoft Mixed Reality. The Analysis Platform team has invested several years of engineering work in building internal infrastructure and now we are making these efforts available to the community as open source as part of the Azure Machine Learning ecosystem. The RAI tooling team consists of Ilya Matiach, Mehrnoosh Sameki, Roman Lutz, Richard Edgar, Hyemi Song, Minsoo Thigpen, and Anup Shirgaonkar. They are passionate about democratizing Responsible AI and have several years of experience in shipping such tools for the community with previous examples on FairLearn, InterpretML Dashboard etc. We also received generous help and expertise along the way from our partners at Microsoft Aether Committee and Microsoft Mixed Reality: Parham Mohadjer, Paul Koch, Xavier Fernandes, and Juan Lema. All marketing initiatives, including the presentation of this blog, were coordinated by Thuy Nguyen.

 

Big thanks to everyone who made this possible!

 

Related research

Towards Accountable AI: Hybrid Human-Machine Analyses for Characterizing System Failure. Besmira Nushi, Ece Kamar, Eric Horvitz; HCOMP 2018. pdf

 

Software Engineering for Machine Learning: A Case Study. Saleema Amershi, Andrew Begel, Christian Bird, Rob DeLine, Harald Gall, Ece Kamar, Nachiappan Nagappan, Besmira Nushi, Thomas Zimmermann; ICSE 2019. pdf

 

Updates in Human-AI Teams: Understanding and Addressing the Performance/Compatibility Tradeoff. Gagan Bansal, Besmira Nushi, Ece Kamar, Daniel S Weld, Walter S Lasecki, Eric Horvitz; AAAI 2019. pdf

 

An Empirical Analysis of Backward Compatibility in Machine Learning Systems. Megha Srivastava, Besmira Nushi, Ece Kamar, Shital Shah, Eric Horvitz; KDD 2020. pdf

 

Understanding Failures of Deep Networks via Robust Feature Extraction. Sahil Singla, Besmira Nushi, Shital Shah, Ece Kamar, Eric Horvitz. arXiv 2020. pdf

Adding Custom Metadata Through a Content Pack in Learning Pathways

by Contributed | Feb 18, 2021 | Technology

This article is contributed. See the original author and article here.

Overview 

Learning Pathways is a customizable, on-demand learning solution in Microsoft 365. Learning Pathways offers a customizable SharePoint Online communication site (that may be used as a training portal), easy provisioning from the SharePoint Look Book, the ability to make your own training playlists with custom assets, a custom web part to surface training content across your SharePoint site collections, and up-to-date Microsoft documentation for Microsoft 365 solutions. 

The information architecture behind Learning Pathways supports structuring your playlists by category and subcategory. Within a playlist, you may add custom assets (in the form of SharePoint site pages) or use the content provided by Microsoft. For each playlist, you may add additional information for the playlist title, description, technology, subcategory, level, and audience. While you may add your own subcategories to Learning Pathways, out of the box you cannot add new categories or choices within technology, level, or audience.  

Some organizations may find the existing choices do not support their needs. To customize these fields, you will need to create a custom content pack within GitHub and add the content pack to your Learning Pathways instance. 

Deciding to Make a Custom Content Pack 

There are two primary reasons an organization may decide to begin using a custom content pack: 

1. Edit the information architecture in Learning Pathways further than what is capable out of the box. As we discussed, there are some fields where you cannot add values. 

1. Control the release of Microsoft’s automatic content updates to Learning Pathways. Some organizations might want to review the new content releases to evaluate what is applicable to the organization before it appears in their Learning Pathways instance. 

Please note, creating a custom content pack also means you will need to submit a pull request from the main Learning Pathways repo to your forked repo to take advantage of any content updates. In layman’s terms, you divided your content from the Learning Pathways original source so you will not get the automatic content updates from Microsoft for new docs.microsoft.com content. You will instead need to complete manual steps to get that content into your custom content pack. 

Pre-Work: Provision Learning Pathways 

Follow the docs.microsoft.com instructions for provisioning Learning Pathways. 

Step 1: Fork the Learning Pathways Repo 

Navigate to https://github.com/pnp/custom-learning-office-365 and click Fork in the upper-right hand of the page. This will create an identical copy of the Learning Pathways content in your own repository allowing you to make customizations to the information architecture through editing the JSON. 

After you are done forking the repo, you will see your own copy of the repo in the top left navigation: 

Step 2: Turn on GitHub pages 

Click Settings in the top navigation: 

 Scroll down the page until you see a header for GitHub Pages. In the Source dropdowns, select Main and /docs then click Save: 

Upon saving, GitHub will bring you to the top of the page again. Scroll down to GitHub Pages once more to copy the URL for your GitHub pages: 

Step 4: Add GitHub pages as a custom content pack to Learning Pathways 

Follow the docs.microsoft.com instructions for adding a content pack to learning pathways. When adding the URL for your custom content pack, paste the URL from step 3, add learningpathways/ to the URL, and click Save. For example: 

https://eemancini.github.io/custom-learning-office-365/learningpathways/ 

This adds your forked copy of Learning Pathways as a tab in the site page CustomLearningAdmin.aspx of Learning Pathways: 

Step 5: Edit metadata in GitHub 

As of this step, your custom content pack is an identical copy of Learning Pathways content as you have not made any edits in the repo. Navigate to https://[yourusername].github.io/custom-learning-office-365/learningpathways/v4 to begin making edits. Open the applicable language folder, in this example we will be working in en-us. In this folder you will find 3 JSON files. Select metadata.json:

Explore the metadata.json structure for more guidance on how to edit the information architecture within this JSON file. If you are new to JSON, Bob German’s Introduction to JSON provides an excellent overview for beginners. Watch an example video of editing the existing technologies field and adding new ones. 

Step 6: Commit changes to main branch of your forked repo 

After completing your JSON edits, scroll to the bottom of the page to the section for Commit Changes. Add a title and description that clarifies what edits you made and click Commit Changes. Once you commit a change, your edits will automatically appear in Learning Pathways. (Note: You may need to hard refresh or clear your cache to see these changes). 

Step 7: Add the web part to a page and filter to the content pack 

Now that your custom content is added to Learning Pathways, you can surface it by adding a Learning Pathways web part to a page. Follow the docs.microsoft.com instructions on how to filter to the content pack. 

Conclusion

Whether you are creating a content pack to customize the metadata or control the content releases by Microsoft, creating a custom content pack is a powerful way to support your needs as long as you can support manually pulling content from the Learning Pathways repo in the future.

Upload Custom Linux Hyper-V Image for Gen 2 VM in Azure

by Contributed | Feb 18, 2021 | Technology

This article is contributed. See the original author and article here.

Introduction

This is Andrew Coughlin and I am a Customer Engineer at Microsoft focusing on Azure IaaS. In this blog I will focus on how to upload a custom linux Hyper-V image for generation 2 virtual machines in Azure. Support for generation 2 virtual machines in Azure was released on November 4, 2019, these systems use UEFI boot, SCSI, supported on VMs that support premium storage compared to generation 1 which use PCAT boot, IDE and are available on all VM sizes.  Generation 2 VMs provide for OS disk size greater than 2TB in size and build larger VMs, up to 12TBs. To get additional information about generation 2 virtual machines in Azure please visit this post.

If you have ever uploaded a custom image in Azure in the past you will notice the process is very similar.  Are you looking for the process for Windows VMs, head over to my post here that covers this same process but for Windows VMs.

Prerequisites

• Review the Support for generation 2 VMs on Azure.


• Install Azure PowerShell as documented here.


• Create Hyper-V Gen 2 machine to be used as the image.


• Review Generic Linux image documentation here.


• Review the specific distribution preparation documentation:
  
  
  
  • Debian is located here.
  
  
  • Oracle Linux is located here.
  
  
  • OpenBSD is located here.
  
  
  • Redhat is located here.
  
  
  • SUSE is located here.
  
  
  • Ubuntu is located here.
  
  
  
  


• Convert VHDX to VHD as documented here.


• Download and Install azcopy as documented here.


• Sign in with Azure as documented here.


• Select Azure subscription to upload image as documented here.

Upload Hyper-V Image to Managed Disk

First you want to determine which resource group the image will reside in or if you will create a new resource group. As a reminder a resource group is a container that holds related solutions: virtual machines, storage accounts, disks virtual networks, etc. for your Azure solutions. A resource group can include all resources for your solutions or only those resources that you want to be managed together. For documentation on how to create a new resource group, this can be found on this page. In this example I’m going to use the resource group called “rg-images”.

First, we need to open an elevated PowerShell command prompt.

Next, we will set some variables as we will need these throughout the process, in this example. We are going to be creating this image in Central US, with the image name CentOS7-Image-V2, in the resource group called rg-images, with a disk image called CentOS7-Image-V2.

• $location = ‘Central US’


• $imageName = ‘Ubuntu1804-Image-V2’


• $rgName = ‘rg-images’


• $diskname = ‘Ubuntu1804-Image-V2’

Next, we want to create an empty managed disk, we will type the following commands:

$vhdSizeBytes = (Get-Item "<full File Path>").length
$diskconfig = New-AzDiskConfig -SkuName 'Standard_LRS' -OsType 'Linux' -UploadSizeInBytes $vhdSizeBytes -Location $location -CreateOption 'Upload'
New-AzDisk -ResourceGroupName $rgName  -DiskName $diskname -Disk $diskconfig

NOTE: You can replace the Standard_LRS, with Premium_LRS or StandardSSD_LRS.  At the writing of this blog Ultra disks are currently not supported.

Next, we need to confirm the disk status is equal to “ReadyToUpload”, we will type the following:

$disk = Get-AzDisk -ResourceGroupName $rgName -DiskName $diskname
$disk.DiskState

NOTE: The disk status must be set to “ReadyToUpload”, if it is not you need to check what was typed in the “New-AzDiskConfig” command.

Now we want to create the writable shared access signature (SAS) for the managed disk we just created.  Then we will get the disk status and make sure it is equal to “ActiveUpload”, to do this we will type the following:

$diskSas = Grant-AzDiskAccess -ResourceGroupName $rgName -DiskName $diskname -DurationInSecond 86400 -Access 'Write'
$disk = Get-AzDisk -ResourceGroupName $rgName -DiskName $diskname
$disk.DiskState

Now we are ready to upload our disk to Azure, to do this we will type the following and wait for the process to complete:

cd <location of Azcopy>
.azcopy.exe copy "<location of vhd>" $diskSas.AccessSAS --blob-type PageBlob

When the upload is completed you will get the following results:

After the upload has completed, we will revoke access from the disk as we no longer need the shared access signature we created above, we will type the following:

Revoke-AzDiskAccess -ResourceGroupName $rgName -DiskName $diskname

Create Image from Managed Disk

We now have the managed disk uploaded to the cloud.  The next step is to create an image from that managed disk.  When the image is created, we want to make sure to specify this image will be a V2 image.  To do this we will type the following:

$imageConfig = New-AzImageConfig -Location $location -HypervGeneration V2
$imageConfig = Set-AzImageOsDisk -Image $imageConfig -OsState Generalized -OsType Linux -ManagedDiskId $disk.Id
$image = New-AzImage -ImageName $imageName -ResourceGroupName $rgName -Image $imageConfig

Verify in the portal that our image is now created from our managed disk.  We can now start provisioning generation 2 virtual machines with this image.

Conclusion

There you have it; we have just uploaded a custom linux image and now we can use that image to deploy generation 2 virtual machines in your Azure environment.  Thank you for taking the time to read this blog, I hope this helps you and see you next time.

Disclaimer

The sample scripts are not supported under any Microsoft standard support program or service. The sample scripts are provided AS IS without warranty of any kind. Microsoft further disclaims all implied warranties including, without limitation, any implied warranties of merchantability or of fitness for a particular purpose. The entire risk arising out of the use or performance of the sample scripts and documentation remains with you. In no event shall Microsoft, its authors, or anyone else involved in the creation, production, or delivery of the scripts be liable for any damages whatsoever (including, without limitation, damages for loss of business profits, business interruption, loss of business information, or other pecuniary loss) arising out of the use of or inability to use the sample scripts or documentation, even if Microsoft has been advised of the possibility of such damages.